bigknight / csrnet_pytorch Goto Github PK

请问一下，在csrnet那篇论文中，最后有psnr和ssim的比较，源网络不是有3次池化吗，那么最后怎么能比较psnr和ssim呢?

hello i don't found test model code in this code ,could you help me .and i have some question about data augment ,crop train data 1/4 size ,but how to do with test data. i am sorry to bother you .and very very thanks.

密度图的ground-truth是根据标签生成的，然后根据密度图积分求和得到人数，对吗？
但是如果用一个随便找的图片进行测试的时候，没有标签，那怎么生成的密度图？
能解答一下吗

I found you have confirmed the experiments in the paper of CSRNet are all right. Have you ever try the WorldExpos' dataset? I cannot reimplement the results for this dataset. If you have done it, please tell me. Thanks very much!

Note that it seems the labels of CSRNet is not right. You should pay attention to it.